Probability, Explanation and Inference: A Reply
| Published date | 01 October 2007 |
| DOI | 10.1350/ijep.2007.11.4.307 |
| Date | 01 October 2007 |
| Subject Matter | Article |
Probability, explanation
and inference: a reply
By Ronald J. Allen* and
Wigmore Professor of Law, Northwestern University; Fellow,
Procedural Law Research Center, China Political Science and Law
University
Michael S. Pardo†
Assistant Professor, University of Alabama School of Law
he inferences drawn from legal evidence may be understood in both
probabilistic and explanatory terms. Consider evidence that a criminal
T defendantconfessedwhileinpolicecustody.Toevaluatethestrengthof
this evidence in supporting the conclusion that the defendant is guilty, one could
try to assess the probability that guilty and innocent persons confess while in
police custody. Or one could make the same assessment based on any number of
other characteristics shared by the defendant or the context of the confession. The
problem of reference classes would arise quite readily because each of these
different classes would likely yield different results, some of which will take one
closer to, and some further away from, the correct conclusion of whether this
defendant in fact is guilty. Alternatively, one could evaluate how well the
conclusion that the defendant is guilty explains the evidence of the confession.1
How well the defendant’s guilt explains the evidence will depend on the strength
of alternative explanations such as whether a false confession was coerced, or the
defendant was trying to protect another person from conviction, or the police are
lying about whether a confession was ever given, etc.
*
Email: rjallen@law.northwestern.edu. Professor Allen is indebted to the Julius Rosenthal
Foundation and the Searle Center for support of the preparation of this article.
†
Email: mpardo@law.ua.edu. My thanks to Dean Ken Randall and the University of Alabama Law
School Foundation for generous research support.
1
Both approaches involve inductive inferences; the probability approach employs enumerative
inferences and the explanation approach employs abductive inferences.
THE INTERNATIONAL JOURNAL OF EVIDENCE & PROOF
(2007) 11 E&P 307–317
307
PROBABILITY, EXPLANATION AND INFERENCE: A REPLY
To be sure, evidence expressed in probabilistic terms may sometimes be helpful in
evaluating the plausibility of completing explanations. We nonetheless contend
that such explanatory considerations better explain important aspects of the
nature of juridical proof, including the value of legal evidence and standards of
proof, than competing probabilistic accounts2 Probability approaches to judicial
evidence and proof, including formal mathematical models, are useful to the
extent that they contribute to these explanatory considerations and less so when
they do not. We elaborate on these points in reply to the four papers in this
symposium. The responses by Nance, Colyvan and Regan, and Rhee, which we
discuss first, focus primarily on the probability perspective. Laudan focuses on the
explanatory perspective, to which we turn in conclusion.
Nance
Nance argues that the reference class problem does not preclude the use of formal
models as tools to evaluate inferences. He suggests that we are unnecessarily
sceptical in rejecting the formal analysis and that those modelling such evidence
are aware of several of the limitations that we pointed out, including that: the
correct reference class cannot be deduced from the environment; class selection
involves judgment informed by inferential interests and epistemic values;
probative value cannot be equated with random or arbitrarily chosen classes; and
the need to consider multiple classes. Significantly, Nance suggests, in such
models there is no ‘claim or implicit assumption that a unique reference class is
necessarily privileged’.3
First, we are not uncompromising sceptics, and we even concede that the formal
models may be useful in evaluating evidence and that it may not be unrea-
sonable for parties to argue, or fact-finders to evaluate evidence, along the lines
suggested by such models. To say that such models may be useful is not, however,
to accept them as the sole or even a particularly reliable method of discovering
the truth. Our objection is to scholarship arguing just that such models establish
the correct or accurate probative value of evidence, and thus implying that any
deviations from such models lead to inaccurate or irrational outcomes. We think
it telling that Nance labels critics of this one methodology as sceptics, as though
there were no other paths to truth and no other tools of rationality. This is a very
narrow view of rationality and inquiry, and we do indeed reject it. Formal
methods of the sort Nance is espousing are powerful but within limited
2
We discuss this conclusion and these issues in more detail in Michael S. Pardo and Ronald J. Allen,
‘Juridical Proof and the Best Explanation’ Law and Philosophy (forthcoming).
3
Nance, ‘The Reference Class Problem and Mathematical Models of Inference’, above at 259.
308
E & P
PROBABILITY, EXPLANATION AND INFERENCE: A REPLY
domains, and most of human experience, and thus most inferential problems,
do not lie within those domains.
Secondly, Nance’s protestations to the contrary, we think it obvious that the
models in the articles that we addressed do privilege certain reference classes,
when such classes and such models are by no means the only reasonable way to
evaluate the evidence. Thus, the strong conclusions drawn from the models are
unwarranted. To demonstrate we focus again on the Nance and Morris DNA
studies and the arguments by Finkelstein and Levin.
Nance responds that the ‘main conclusions in the Nance and Morris studies’ are
‘that jurors tend to undervalue DNA match evidence relative to the stated
Bayesian norms, but that the degree of undervaluation can be reduced by certain
kinds of Bayesian instruction’.4 The authors obviously think that reducing this
‘undervaluation’ is a good thing; they write that ‘a careful use of such Bayesian
methods in the courtroom can indeed assist the jury in reaching more accurate
verdicts’.5 Although Nance asserts that ‘no fair reading’ of the studies takes them
as ‘privileging’ reference classes, nor as declaring juror conclusions to be incorrect
or irrational,6 how can one speak of undervaluing the evidence or reaching more
accurate verdicts without assuming that the models are providing a correct evalu-
ation of the evidence? To assert an error in the experimental responses, the models
must privilege some reference class to calculate the supposed correct response,
whether the random-match...
To continue reading
Request your trial